Symplectic Lefschetz fibrations and the geography of symplectic 4-manifolds
Matt Harvey
January 17, 2003
This paper is a survey of results which have brought techniques from the theory of complex surfaces to bear on symplectic 4-manifolds. Lefschetz fibrations are defined and some basic examples from complex surfaces discussed. Two results on the relationship between admitting a symplectic structure and admitting a Lefschetz fibration are explained. We also review the question of geography: 2 which invariants (c1, c2) occur for symplectic 4-manifolds. Constructions of simply connected symplectic manifolds realizing certain of these pairs are given.
1 Introduction
It is possible to answer questions about a broad class of 4-manifolds by using the fact that they admit a special kind of structure called a Lefschetz fibration: a map from the manifold to a complex curve whose fibers are Riemann surfaces, some of them singular. The interesting data in such a fibration is the number of singular fibers and the types of these singular fibers. The types can be specified by monodromies (elements of the mapping class group of a regular fiber) associated to loops around the critical values of the fibration map in the base curve.
Lefschetz fibrations should, at least in theory, frequently be applicable to ques- tions about symplectic 4-manifolds, in light of a theorem of Gompf that a Lef- schetz fibration on a 4-manifold X whose fibers are nonzero in homology gives a symplectic structure on X and a theorem of Donaldson that, after blowing up a finite number of points, every symplectic 4-manifold admits a Lefschetz fibration.
After giving some examples from the category of complex surfaces and dis- cussing the above results on Lefschetz fibrations, we state and sketch a proof of Gompf’s theorem that fiber sums can be carried out along codimension 2 sym-
1 plectic submanifolds so that the resulting manifold will still carry a symplectic structure.
Next is a brief introduction to the geography problem. The interesting part of this problem in the symplectic category is the question of existence of symplec- tic manifolds which are not homeomorphic to any complex surface. We will see that such manifolds do exist, and in fact all possible invariants of spin symplec- tic manifolds lying below the so-called Noether line are easy to realize using Gompf’s symplectic fiber sum.
There are a few conventions used throughout the paper. The word curve always means complex curve (dimR = 2), but surface means closed 2-manifold. A complex surface will always be explicitly referred to as such. Coefficients for homology and cohomology are taken in Z when not otherwise specified. The Euler class e(ξ) of a principal U(1) = SO(2) bundle ξ is the same as the first Chern class. We may evaluate e(ξ)[S2] for such a bundle over the 2-sphere and speak of the Euler number. This can be viewed in terms of building total space of the associated disc bundle to the given principal bundle from two copies of D2 × D2. In particular, in the first factor of this product, the construction is that of S2 as the union of the “northern” and “southern” hemispheres along the “equator.” The fibers in the second factor are identified by assigning an element of SO(2) to each point of the equator. By a diffeomorphism, we can choose the element of SO(2) assigned to some fixed base point on the equator to be the identity. The identifications over the equator then give a map S1 → SO(2). The isomorphism type of the principal bundle depends only on the homotopy class ∼ of this map, which is an element of π1(SO(2)) = Z called the Euler number of the bundle. PD(·) is used in various contexts to denote the Poincar´edulaity ∼ n−k isomorphism Hk(X) = H (X) for compact manifolds. We sometimes abuse notation and identify the latter group with the deRham cohomology.
2 Building blocks in the complex category
The following definition will be needed later, but it also provides one of the most basic examples of a 4-manifold admitting a Lefschetz fibration. (This particular one has no critical points and no singular fibers.)
Definition 2.1 A ruled surface is a compact complex surface S with a map 1 π : S → C to a complex curve C such that every fiber is a complex line CP .
2 2.1 Lefschetz fibrations
The structure of Lefschetz fibrations comes from a decomposition of complex surfaces called a Lefschetz pencil.
n If S ⊂ CP is a complex surface, then a generic codimension 2 linear subspace n A ⊂ CP meets S in some finite set of points B. We may then consider two n codimension 1 linear subspaces of CP which both contain the subspace A but otherwise are generic. We can specify these subspaces as the zero sets of two linear homogeneous polynomials V (p0) and V (p1).
1 Now for fixed t = [t0 : t1] ∈ CP , consider the variety Lt = V (t0p0 + t1p1). This n will be a codimension 1 linear subspace of CP containing A. By a dimension count, we see that Lt intersects our original surface S in a complex curve, 1 which may be singular. Choosing s, t ∈ CP , note that Ls ∩ Lt = A so that (Ls ∩ S) ∩ (Lt ∩ S) = B. That is, all of the curves obtained above intersect exactly in the finite set of points B. After blowing up each point of B, we obtain 2 1 1 a well-defined map X#|B|CP → CP by sending each point to that t ∈ CP specifying the curve Lt ∩ S on which it lies.
Definition 2.2 A Lefschetz fibration on a 4-manifold X is a map π : X → Σ where Σ is a closed 2-manifold having the following properties:
1. The critical points of π are isolated.
2. If P ∈ X is a critical point of π then there are local coordinates (z1, z2) on X and z on Σ with P = (0, 0) and such that in these coordinates π is 2 2 given by the complex map z = π(z1, z2) = z1 + z2 .
In terms of this descrtiption of the critical points, a Lefschetz fibration can be thought of as a kind of complex analogue of a Morse function.
Note that if the 4-manifold X is closed (∂X = ∅ and X is compact) then the generic fiber F of π is a closed 2-manifold, so a Riemann surface of some genus. Part of the utility of Lefschetz fibrations is that this simple property means that X can be reconstructued once we know how the fibers are identified if one travels around some loop in Σ \ C where C ⊂ Σ is the set of critical values of π. That is, Lefschetz fibrations admit a combinatorial description in terms of their monodromy representation ρ : π1(Σ \ C) → π0(Diff(F )) from the fundamental group of Σ \ C into the group of homotopy classes of self-diffeomorphisms of F . The latter group is often called the mapping class group.
3 2.2 Examples
2 2 A very simple example of a Lefschetz fibration is a map from CP #CP → 1 2 CP . One begins with a point P ∈ CP and two generic lines containing this point, which we can take to be the vanishing sets V (p0) and V (p1) of two 2 independent degree 1 polynomials. For any point Q ∈ CP \ P there is a unique line V (pQ) containing both P and Q given as the vanishing set of another degree 1 polynomial. Some easy linear algebra shows that one can write pQ = t0p0+t1p1 1 for some [t0 : t1] ∈ CP . The map Q 7→ [t0 : t1] gives the desired fibration on 2 CP \ P . Since we can view blowing up the point P as replacing the point P by the set of all lines passing through P , we obtain a well-defined fibration 2 2 1 2 2 CP #CP → CP . Note that this shows CP #CP is a ruled surface since the 1 fiber over a point [t0 : t1] is V (t0p0 + t1p1), a CP .
2 One can play a similar game starting with points P1,...,P9 ∈ CP given as the intersection V (p0)∩V (p1), where p0 and p1 are two generic homogeneous cubics in three variables. The nine intersection points are in general position. We will be interested in vanishing sets of polynomials of the form
3 3 3 2 2 a1X + a2Y + a3Z + a4X Y + a5X Z+ 2 2 2 2 a6Y Z + a7Z X + a8Z Y + a9Y X + a10XYZ.
2 Picking a point Q ∈ CP \{P1,...,P9} and asking that its coordinates [X : Y : Z] satisfy the above equation gives one linear relation on the ai. In the same way P1,...,P9 give 9 more. Generically, we obtain in this way a homogeneous cubic which (by more linear algebra) can be written as pQ = t0p0 + t1p1. Note 0 that all points Q ∈ V (pQ) give the same [t0 : t1]. Again, by viewing the blow-up of each of the Pi as replacing that point by the set of all points passing through 2 2 1 it, we obtain a map CP #9CP → CP whose generic fibers are elliptic curves 2 in CP , which are topologically the two-dimensional torus T 2. A holomorphic map π : S → C from a complex surface to a complex curve whose generic fibers are nonsingular elliptic curves is called an elliptic fibration. The above is an example known as E(1). There is also the notion of C∞ elliptic fibration which is roughly a map from a 4-manifold to a 2-manifold which is locally diffeomorphic to a holomorphic elliptic fibration by diffeomorphisms of open subsets of S and C commuting with the given maps S → C.
These two examples help one picture the more general construction. Given a holomorphic line bundle L → X over a complex manifold X with two sections σ0, σ1 ∈ Γ(X; L) which are transverse to the zero section, one can consider the 1 1 vanishing of t0σ0 + t1σ1 where [t0 : t1] ∈ CP to get a map X \ A → CP , where A = V (σ0) ∩ V (σ1).
4 2.3 Fiber sum
There is a way to piece together two Lefschetz fibrations π1 : X1 → Σ1 and π2 : X2 → Σ2 whose generic fibers have the same genus to obtain a new Lefschetz fibration π : X1#φX2 → Σ1#Σ2.#φ denotes the fiber sum that is about to be defined and # the ordinary connected sum of the base spaces. One begins with neighborhoods νi of generic fibers Fi of πi for i = 1, 2. These are diffeomorphic 2 to D ×Σg where Σg is the Riemann surface with the same genus as the Fi. One then picks a diffeomorphism φ : S1×Σg (orientation-reversing) of the boundaries of Xi \ νFi and identifies them via φ.
A special case that should be mentioned here is E(2) = E(1)#φE(1) with φ : S1 ×T 2 → S1 ×T 2 the identity map. Topologically this is a K3 (or Kummer) 3 surface, homeomorphic to a degree 4 hypersurface in CP . More generally we may inductively form the elliptic surface E(n) = E(n − 1)#φE(1) with φ the identity as in the construction of E(2). We will return to these examples in the discussion of geography.
3 Symplectic versions of these constructions
3.1 Lefschetz fibrations on symplectic manifolds
Symplectic manifolds (X, ω) for which ω ∈ H2(X; R) is an integral class ad- mit topological Lefschetz pencils via the construction for complex manifolds described in the previous section. That is, they are unions of the vanishing sets of ratios of two sections of a certain line bundle over X as the ratios vary 1 over CP . These real codimension 2 submanifolds all intersect exactly in the common zero set A of the two chosen sections. The set A has real codimension 4.
Theorem 3.1 (Donaldson) If (V, ω) is a symplectic manifold with [ω] integral then for large k, V admits a topological Lefschetz pencil, in which the fibers are symplectic subvarieties representing the Poincar´edual of k[ω].
To obtain the structure of a Lefschetz fibration, we will need to blow up X along this set A, which for a 4-manifold is just a finite set of points. Some remarks on the proof of the above theorem follow. First note that the requirement that ω be an integral class (which seems to show up as a hypothesis of many other results in addition to this one) is not so severe.
Proposition 3.1 [Gom95] Any closed symplectic manifold (M, ω) admits an- other symplectic form ω0 whose cohomology class is in the image of H2(M, Z) →
5 2 HDR(M).
Proof. Fix a metric on M and let B be the -ball about 0 in the space of harmonic1 2-forms on M. Since nondegeneracy is an open condition, all of the forms in ω + B will be symplectic when is sufficiently small. Since ω + B 2 2 covers an open subset of HDR(M) it contains an element of H (M; Q), which 0 we may multiply by a suitable integer to get the desired ω . Holomorphic sections were used to pick out the fibers in the construction of a Lefschetz fibration on a complex manifold, so all the fibers also have the structure of complex manifolds. One can pick a compatible almost complex structure J on any symplectic manifold (X, ω), but there is no guarantee that there will be any complex line bundle with a sufficient supply of J-holomorphic sections to construct the Lefschetz pencil. Donaldson remedies this defect by considering an appropriate line bundle with sections which are “approximately holomorphic.”
ω Since ω is an integral class, we can form a complex line bundle L with c1(L) = 2π and pick a unitary connection on L with curvature −iω. This lets one define a ∂¯ operator on the sections of L.
Theorem 3.2 [Don96] Let L → X be a complex line bundle over a compact symplectic manifold (X, ω) with compatible almost-complex structure, and with ω c1(L) = 2π . Then there is a constant C such that for all large k there is a section s of L⊗k with C |∂s¯ | < √ |∂s| k on the zero set of s. Where ∂ and ∂¯ are the complex linear and anti-linear parts of the derivative ∇ which is canonically defined along the vanishing of any section of L.
We can apply the following proposition on each tangent space to the vanishing set of a section to see that for a section s ∈ Γ(X; L⊗k) satisfying |∂s¯ | < |∂s| we have s−1(0) ⊂ M a symplectic submanifold.
Proposition 3.2 Suppose that A : Cn → C is a real linear map. Let a0 and a00 be the complex linear and conjugate linear parts of A. Then if |a00| < |a0| then ker(A) is symplectic with respect to the standard symplectic structure ω(v, w) = Im hv, wi, where h·, ·i is the standard Hermitian metric on Cn. 1This means those η ∈ Ω2(M) such that 4η = (dd∗ + d∗d)(η) = 0 where d is exterior differentiation and d∗ is its adjoint with respect to the bilinear forms given on Ω∗(M) by the metric. We view d, d∗ :Ωeven(M) → Ωodd(M). It is a theorem that every deRham cohomology class has a unique harmonic representative.
6 3.2 Symplectic structure on Lefschetz fibrations
The converse to the result in the last section is considerably easier to prove. The argument below follows [GS99].
Theorem 3.3 Suppose that X is a closed 4-manifold with a Lefschetz fibration π : X → Σ. Denote the homology class of a regular fiber by [F ] ∈ H2(X). Then X admits a symplectic structure which is symplectic on all the fibers of π if and only if [F ] 6= 0.
Proof. [F ] 6= 0 is necessary, since given a symplectic form ω ∈ Ω2(X) which is R also symplectic on fibers we would have F ω > 0 for F a regular fiber.
2 R If [F ] 6= 0 then there exists some α ∈ Ω (X) with F α > 0. While this is a big step towards constructing a symplectic form on X which is also symplectic R on the fibers, we have to make sure that α > 0 for each closed surface F0 F0 contained in a singular fiber S, which is topologically a regular fiber with some loops called vanishing cycles collapsed to points.
Singular fibers are still homologous to regular fibers. If S = π−1(c) is the preimage of a critical value c ∈ Σ of π, and F = π−1(d) is a regular fiber, then considering the preimage of an arc from c to d in Σ shows that [S] = [F ] ∈ H2(X). The same argument shows that all the regular fibers are homologous.
If a singular fiber S is obtained from F by collapsing a separating vanishing R cycle, then we have S = F0 ∪ F1 for some closed surfaces F0 and F1 and F α = R α + R α > 0, so if R α = r ≤ 0 then s = R α > 0. We can replace α by F0 F1 F0 F1 0 s R R 0 s α = α+(−r+ )PD(F1). PD(F1) = [F0]·[F1] = 1, so α = r−r+ > 0 2 F0 F0 2 R 0 s and α = s + (−r + )[F1] · [F1] > 0. All other fibers F of π do not intersect F1 2 R 0 R F1 and so have F α = F α. Since there are only finitely many isolated critical points of π we obtain, after finitely many modifications of this kind, a closed 2 R form ζ ∈ Ω (M) such that ζ > 0 for each closed surface F0 contained in a F0 fiber of π.
Now we need to fix this to be symplectic on fibers. Pick disjoint open balls Uj ⊂ X about each critical point of π so that on each Uj there are local coordinates 2 2 with in which we have π(z1, z2) = z1 + z2 . We have the standard symplectic structure ωUj = dx1 ∧ dy1 + dx2 ∧ dy2 on each Uj. π is holomorphic, so if y ∈ Σ −1 2 then π (y) ∩ Uj is a holomorphic curve, which means it is ωUj -symplectic. −1 For fixed y, extend the symplectic structure on π (y) ∩ ωUj to a symplectic −1 −1 structure ωy on all of π (y). (Extend means pick an ωy if π (y) ∩ Uj = ∅.) R R We can rescale each ωy staying away from ∪Uj to make F ωy = F ζ so that 2 [ωy] = [ζ] ∈ HDR(X; R) for all y. Now we need to glue these together. 2In general if J is an ω-compatible almost complex structure then C is J-holomorphic if and only if ω|C is symplectic.
7 Pick a finite cover {Wj} of Σ with at most one critical value in each Wj. Pick yj ∈ Wj for each j, choosing yj to be the critical value for those Wj which −1 contain a critical value. If π (Wj) is just a collection of regular fibers, let −1 −1 −1 rj : π (Wj) → π (yj) be a retraction. If π (yj) is a singular fiber, then take −1 −1 3 a retraction rj : π (Wj) → π (y) ∪ cl(Uj). Choose also a partition of unity P 2 ρj = 1 subordinate to this cover. We can define forms ηj ∈ Ω (Wj) for each j by ∗ r ωUj (P ), r(P ) ∈ cl(Uj) ηj(P ) = ∗ r ωyj (P ), otherwise. 1 −1 Since we made the [ωy] = [ζ] there exist θj ∈ Ω (π (Wj)) so that
−1 (ωy − ζ) |π (Wj ) = dθj for all j. We can then form
X 2 η = ζ + d (ρj ◦ π)θj ∈ Ω (X). P This has dη = dζ = 0 and so is a closed form. For a given fiber ηFy = ρj(y)ηj is linear combination of forms sympelctic on Fy with coefficients in [0, 1] and so is itself symplectic on Fy. To get a symplectic form on X, let ∗ ωt = tη + π ωΣ, where ωΣ is any symplectic form you choose on the base. ωt is closed for all t since η and ωΣ are closed. Along fibers, ωt agrees with tη and so is nondegen- −1 erate. If x ∈ π (y) is not a critical point of π then for v, w ∈ TxFy we have ∗ ωt(v, w) = tη(v, w) since π ωΣ is zero along fibers. This also shows that we have ωt η equal symplectic orthogonals (TxFy) = (TxFy) ⊂ TxX. Since η is symplectic ωt ∗ on Fy we have TxX = TxFy ⊕ (TxFy) , and π ωΣ is nondegenerate on the second factor. Since nondegeneracy is an open condition, ωt is nondegenerate on TxX for small enough t.
The argument above worked away from the critical points of π. In particular, since X\∪Uj is comapct, there is some t0 with ωt symplectic for 0 < t ≤ t0. Now ∗ on the Uj we have (ωt)|Uj = tωUj +π ωΣ. Recall that on the Uj we have complex structures on base and total space such that π is a holomorphic map. We can choose ωΣ so that the almost complex structure on the base is compatible, and we have ωUj compatible with multiplication by i in Uj by definition of ωUj . If v ∈ TUj is given in the local coordinates (z1 = x1 + iy1, z2 = x2 + iy2) by v = a ∂ + b ∂ + c ∂ + d ∂ then ∂x1 ∂y1 ∂x2 ∂y2
∗ ωt(v, iv) = tωUj (v, iv) + π ωΣ(v, iv)
= t(dx1 ∧ dy1 + dx2 ∧ dy2)(v, iv) + ωΣ(π∗v, π∗(iv)) 2 2 2 2 = t(a + b + c + d ) + ωΣ(π∗v, iπ∗v) > 0 3cl(·) means the closure.
8 because it is the sum of two positive terms, so all the ωt are nondegenerate on ∪Uj as well as on its complement, so we have the desired symplectic form on X with all fibers symplectic and see that it arises as a perturbation of the pullback ∗ π ωΣ of a symplectic form on the base.
3.3 Blowing up
Blowing up is a familiar procedure for removing non-transverse intersections and self-intersections of curves. One views this locally at P ∈ C2 by picking a neighborhood of P diffeomorphic to C2 in such a way that P = (0, 0). Then let
2 1 γ = {(Q, l) ∈ C × CP | Q ∈ l}.